| Summary: | <p>Metagenomic sequencing is heralded as a transformative tool for diagnostic clinical microbiology, offering expedited species detection and antimicrobial resistance (AMR) prediction. This thesis introduces a real-time sequencing pipeline direct-from-blood cultures, aiming to replace traditional culture-based diagnostics and antimicrobial susceptibility testing (AST) for suspected bloodstream infections. </p>
<p>I assessed DNA extraction kits and human DNA depletion methods for optimal direct-from-blood culture sequencing. To address potential sequencing errors and taxonomic misclassifications, I systematically benchmarked prevalent classification tools, databases, and sequencing platforms using a combination of simulated and real-world datasets. I processed 271 liquid blood culture samples (211 culture-positive and 62 culture-negative) in real-time, and applied a threshold-based and machine learning approach for species classification. </p>
<p>For species, the initial diagnostic sensitivity was 97% (95 CI: 94-99%, 210/216) and specificity of 94% (90-97%, 193/205) in culture positive samples. After adjusting for plausible false positive samples, the pipeline had a sensitivity of 100% (98-100%, 229/229) and specificity of 100% (98-100%, 205/205). Additionally, 13 polymicrobial and 5 monomicrobial infections that were overlooked by standard routine procedures were identified. In the first ten minutes of sequencing, using re-trained filter thresholds, the sensitivity and specificity was 98% (95-99%, 224/229) and 100% (98-100%, 205/205) respectively, taking a total time of 3hrs30min to species result, nearly half the time (i.e., 10hrs33min) taken by the routine microbiology laboratory. Specificity in culture negative samples, after accounting for a single plausible infection detected, was 100% (94-100%, 61/61) across all timepoints.</p>
<p>For AMR prediction, ResFinder after 24 hours, using raw sequencing reads, had the highest overall sensitivity and specificity of 91% (87-94%, 204/224 drug-organism combinations) and 90% (88-91%, 923/1029) respectively across the top 10 clinically relevant species, while raw reads after two hours of sequencing resulted in a performance of 88% (82-92%, 196/224) and 93% (91-94%, 953/1029), respectively. Notably, this approach allowed for comprehensive AMR detection within 5hrs30min from positive blood culture, outpacing conventional microbiology methods by 22 hours for comparable samples. This study indicates that clinical metagenomics may be integrated into routine diagnostics, allowing for quicker, unbiased pathogen detection, including unculturable and overlooked polymicrobial organisms. This method not only promises prompt antimicrobial resistance predictions but also could bolster patient outcomes. Moreover, it introduces the possibility of bacterial strain typing, vital for infection prevention and oversight.</p>
|
|---|